APL, acute promyelocytic leukemia; ATRA, all-trans retinoic acid; DNMT, DNA methyltransferases; HDAC, histone deacetylases; SGE, stochastic gene expression.
This article suggests a new explanation for the lack of effectiveness of the several differentiation therapies of cancer currently being experimented upon. I propose that the use of molecules targeting chromatin remodelers constitutes the correct strategy, but that it is not based on the right cancer and cell differentiation theories. Instead, a paradigm shift is needed to make it effective. The design of differentiation therapy is always based on deterministic theories of cell differentiation, which state that the linking of stereospecific molecules onto promoters is sufficient to unequivocally induce gene expression and provoke durable differentiation. However, it misses crucial aspects that can only be understood if stochastic gene expression (SGE) and its stabilization during cell differentiation are taken into account. It might not be enough to re-express genes involved in differentiation with molecules such as chromatin remodelers. This is because their expression remains stochastic if no microenvironmental control stabilizes them, risking counter-selection due to their negative effect on proliferation. Based on a stabilization model of cell differentiation, the new therapeutic strategy envisioned here proposes to couple chromatin remodelers with the microenvironmental stabilization of the re-expressed genes, particularly those involved in cellular interactions, to convert malignant cells into benign cells.
Converting malignant cancer cells into benign cells by differentiation is not a new idea. Pierce 1 introduced it after having observed that some malignant cells from teratocarcinoma spontaneously differentiated into benign cells in vivo 2. Subsequently, this differentiation was directly provoked in vitro 3. In the 1970s, important developments came from experiments showing that teratocarcinoma cells are able to generate normal development when transplanted into blastocysts 4.
Later, Sachs 5 studied the suppression of malignant features of leukemic cells by differentiation. Natural molecules involved in normal haematopoiesis were able to differentiate chronic myeloid leukemia cells 5, but only for a sub-population called D+. The other part of the population (D−) was insensitive 6. This work showed for the first time that karyotypically abnormal cancer cells differentiated in vitro (contrary to teratocarcinoma cells, which possess an apparently normal karyotype) 7. Other efforts on leukemia in the 1980s have, so far, provided the sole clinical treatment that was clearly based on differentiation; namely for a subtype called acute promyelocytic leukemia (APL) 8. To treat APL, differentiation is induced by degradation of the chimeric oncoprotein PML/RARα when all-trans retinoic acid (ATRA) is applied. Thus, genes involved in blood cell maturation are re-expressed and enable recovery 9. Unfortunately, ATRA is not effective on solid cancers, despite a slight ability to induce their differentiation 10.
Cancer cells have lost, at least in part, the molecular characteristics of their tissue of origin. But whether they are lost because undifferentiated cells have been selected or because this loss is a side-effect of enhanced proliferation remains an open question 11. As fusion of cancer cells with normal cells suppresses malignancy 12, Harris considered that cell fusion shows that normalization is possible if the differentiation ‘program’ of a normal cell is imposed upon the cancer cell. Since differentiation genes are repressed during cancer progression (because they act as tumour suppressors), their loss is probably a critical step 11. Harris concluded that cancer has to be primarily considered as a differentiation disease (as opposed to proliferative). Nevertheless, Harris 13 stood behind the idea of a genetic origin for these differentiation defects.
Recently, the cancer stem cell concept 10 and the novel molecules targeting chromatin remodelers have caused a renewed interest in differentiation therapy. In the current strategy, one aims at inhibiting chromatin remodelers that close chromatin, in order to allow for the re-expression of differentiation genes and provoke differentiation. For this purpose, histone deacetylases (HDAC) and DNA methyltransferases (DNMT) are promising therapeutic targets. HDAC inhibitors and DNMT inhibitors are being intensively tested in clinical trials 14. Some of them have been approved for some leukemia (azacitidine) or lymphoma (suberoylanilide hydroxamic acid), and their combination together, or with ATRA, might be more effective 15. However, their effectiveness remains limited. Despite their ability to differentiate many cells from solid cancers in vitro 16, they are not effective in the majority of patients 17. Several hypotheses have been advanced to explain this 17, especially the fact that patients with advanced cancer who have already received chemotherapy may have already developed multiple resistance to drugs. Alternatively, they may possess a subset of cells in which random genetic changes impinge upon the ability to reactivate differentiation programs. Moreover, non-genetic heterogeneity is likely to stochastically generate populations resistant to these as to other drugs 18. Similarly, oncogenes contribute to transformation in only a subset of cells that are in a ‘permissive’ state, which appears stochastically and is linked to differentiation. Therefore, this heterogeneity has to be suppressed. Only a therapeutic strategy based on a model taking into account the stochastic nature of gene expression and its stabilization during differentiation by spatial relationships between cells might reach this goal.
I have previously proposed that differentiation defects in cancer cells are caused by a disruption of the cell-cell interactions that normally stabilize SGE and differentiation features 19. Numerous arguments suggest that, during development, cells are primarily and intrinsically unstable because of unrestricted SGE (this is confirmed by the widespread, pervasive and dynamic gene expression in stem cells 20). Subsequently, the cells are canalized towards differentiation by cellular interactions. In this process, and thanks to epigenetic modifications of the proteins present on the promoters when cells interact, the signal transduction generated by cellular interactions stabilizes patterns of gene expression 21, 22. This enables these interactions and the corresponding differentiated states.
In support of this model, gene expression is actually stabilized by tissue structure and cellular interactions 23–25. Because it is based on stabilization rather than induction of gene expression, as is the case in deterministic models, it facilitates predictions that are relevant to the design of a differentiation therapy of cancer. If differentiation is a global decrease in stochasticity of gene expression caused by cellular interactions, disruption of these interactions by any physical, chemical or physiological factor can lead to cancer. In such cases, reacquisition of high SGE by differentiated cells, or failure of the canalization of progenitor cells, is likely to generate cells exhibiting confused and unstable gene expression profiles. Genetic and epigenetic instabilities are also generated because of abnormal expression of DNA repair or chromatin remodelling genes 19. Consequently, it is necessary for a differentiation therapy to restore normal or ‘pseudo-normal’ cellular interactions in order to stabilize gene expression patterns, and ‘recanalize’ cells towards differentiation.
Current differentiation therapies are all based on deterministic models of cell differentiation. They suppose that derepression of differentiation genes is sufficient to generate differentiation. In the stabilization model, this derepression, although necessary, is not sufficient because gene expression will remain stochastic unless signals corresponding to correct interactions with the cellular microenvironment stabilize it (Fig. 1). If these signals are not brought to the cells to stabilize the re-expressed differentiation genes, expression fluctuations will continue and high levels of expression can be counter-selected because of their negative effect on proliferation. Thus, providing cancer cells with the necessary elements from the microenvironment for this stabilization to occur could be the ‘key’.
To test the hypothesis that the cancerous state is not linked to the appearance of a few oncogenes but to the disappearance of complete differentiation, Harris 11 forced the expression of keratin-1, a keratinocyte marker involved in cellular interactions, in skin cancer cells, in order to make them differentiate and behave normally. Their ability to form tumours in mice decreased, but was not completely abolished. Importantly, cells able to form new cancers no longer expressed keratin-1. Its expression had been counter-selected. Harris also showed that, in non-tumorigenic hybrid cells obtained by fusion of normal and cancer cells, artificial suppression of fibronectin expression, a protein involved in cell adhesion to the extracellular matrix, induces a new cancerous state 26. These results support the notion that cancer phenotypes are primarily linked to the absence of adhesion and cellular interactions, in agreement with the stabilization model.
This also shows that when a gene is directly re-expressed thanks to its reintroduction by genetic engineering, or indirectly by removal of its inhibition, cancer cells are still heterogeneous in the gene's expression pattern. This fact can hardly be understood in deterministic models in which all cells are supposed to behave identically when they receive the same stimulus or treatment. On the contrary, it is coherent with the stabilization model: the heterogeneity arises from stochastic fluctuations in each cell and it could only be suppressed if stabilization of gene expression occurs along with re-expression of differentiation genes.
For D+ and D− myeloid leukemic cells, in which differentiation can or cannot be induced respectively 5, Leo Sachs has shown that the difference between these cells is due to the equilibrium between the expression of genes enabling differentiation and those preventing it. Most importantly, he showed that changes in their expression level can suppress a cancerous state by restoring the ability of the cells to differentiate into non-proliferating cells in vivo at some places in the body where they are normally exposed to ‘differentiation inducers’ 5. Thus, interactions with the microenvironment can stop proliferation in vivo, but only if cells express the genes enabling these interactions.
In the stabilization model, gene expression remains highly stochastic if not controlled by the microenvironment; still producing cells with patterns of expression allowing proliferation that are positively selected 19. Thus, two steps have to be achieved to stop proliferation: both the expression of differentiation genes and the stabilization of cellular phenotypes by the presence of molecules able to interact with cancer cells in their microenvironment (Fig. 1).
The first step can be achieved by chromatin remodeler inhibitors (or other ‘derepressing’ molecules) that derepress genes important for differentiation and cellular interactions by suppressing epigenetic modifications that had been selected for in cancer progression. But with this treatment alone, cells become ‘pseudo-stem cells’ that stochastically express genes involved in differentiation, without having the microenvironmental cues to stop their phenotypic fluctuations. This treatment makes reintegration in tissue possible, but it is not enough by itself. The second step, parallel stabilization of cellular phenotypes, is necessary. Since a ‘normal’ microenvironment is not present in tumours, it could be the role of therapeutic intervention to provide it by supplying proteins normally involved in cellular interactions. These proteins could be directly introduced in soluble form to make cells interact with them in order to maintain the expression of differentiation markers. This would stop proliferation and generate stable differentiation of the ‘pseudo-stem cells’. As these ‘artificially-provoked’ interactions would stop phenotypic fluctuations, one could expect that escapes would be very unlikely, especially because genetic and epigenetic instabilities would be ruled out 19. Moreover, potential ineffectiveness due to genetic alterations in genes involved in cellular interactions could be counteracted by using several kinds of interactions and creating an interaction network that also controls cells not expressing functional forms of some cellular interaction proteins.
Alternatively, molecules that are able to actively maintain the expression of genes involved in cellular interactions at a high level could be useful. Recently, differentiation of breast cancer stem cells has been obtained by molecules that incidentally inhibit cancer growth. Interestingly, differentiation is accompanied by an increase in the expression of at least one of the proteins involved in interactions of mammary epithelial cells: E-cadherin 27. The molecules discovered here might directly stabilize the expression of this type of gene and enable phenotype stabilization, differentiation, and arrest of proliferation.
In summary, the limitations of differentiation therapy stem from a lack of understanding of the mechanism of cell differentiation. A paradigm shift is needed and the deterministic theory based on cell induction should be replaced by a stochastic theory based on cell stabilization. When chromatin remodeler inhibitors, for instance, are applied, re-expression of genes involved in differentiation, particularly those involved in cellular interactions and communications, has to be rapidly stabilized. This stabilization could be brought about by cellular interactions mediated by microenvironmental molecules that could be intentionally introduced, even in soluble form, to mimic the normal microenvironment. Only this modification of the current strategy of differentiation therapy could produce long-term recovery by converting cancer cells into stable non-cancerous cells.